a python-based layout-aware analog design methodology for
TRANSCRIPT
HAL Id: hal-00749906https://hal.archives-ouvertes.fr/hal-00749906
Submitted on 8 Nov 2012
HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come fromteaching and research institutions in France orabroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, estdestinée au dépôt et à la diffusion de documentsscientifiques de niveau recherche, publiés ou non,émanant des établissements d’enseignement et derecherche français ou étrangers, des laboratoirespublics ou privés.
A Python-Based Layout-Aware Analog DesignMethodology For Nanometric Technologies
Stephanie Youssef, Farakh Javid, Damien Dupuis, Ramy Iskander,Marie-Minerve Louërat
To cite this version:Stephanie Youssef, Farakh Javid, Damien Dupuis, Ramy Iskander, Marie-Minerve Louërat. A Python-Based Layout-Aware Analog Design Methodology For Nanometric Technologies. IEEE 6th Interna-tional Design and Test Workshop (IDT), Dec 2011, Beyrouth, Lebanon. pp.62-67. �hal-00749906�
A Python-Based Layout-Aware Analog Design
Methodology For Nanometric Technologies.
Stephanie Youssef, Farakh Javid, Damien Dupuis, Ramy Iskander and Marie-Minerve Louerat
University Pierre et Marie Curie (UPMC), LIP6 Laboratory, 4, Place Jussieu, 75005 Paris, France
Email : [email protected]
Abstract—This paper presents a methodology for procedurallayout-aware design for nanometric technologies. A Python-basedlayout generation tool generates different layout styles for thesame basic analog building blocks. Moreover, layout dependentparasitic parameters such as stress effects are easily computedand compared for different layout styles. The procedural layoutdescription is written using a Python API that ensures layoutportability over different technologies. A main focus is on howthe layout generation tool addresses both geometric and parasitic-aware electrical synthesis. This is made possible through aninternal loop that links circularly both the sizing phase and thelayout generation phase. The proposed design methodology assiststhe analog designer in exploring electrical and physical trade-offs.At the end, we present synthesis and characterization results thatprove the effectiveness and speed of the proposed methodology.
I. INTRODUCTION
The demanding of low-power design and a reduced time
to market are the key of success for microelectronics industry.
There are many CAD tools in the digital domain that continue
to improve and ensure the security of the future of the digital
designers. In analog domain this is more complex. The main
obstacle is that unlike the digital circuits, analog circuits are
very sensitive to small variations and the performance may be
affected easily. This is why most of analog designers tradi-
tionally size manually their circuits and perform a full custom
layout which is laborious and error-prone. Another important
point is that with the migration to deep-sub-micron (DSM)
technologies, providing less area, less power consumption
and higher speed integrated circuits, two important constraints
related to the layout of the circuit have to be taken into
consideration:
a) The problem of analog device matching: Due to the
manufacturing process, it is hard to accurately control large
transistors’ widths. Transistor folding technique is commonly
used to reduce parasitic capacitance and gate resistance [1], [2]
allowing more accurate geometries and providing better elec-
trical performance. Interdigitation and mirror styles are usually
used to equally distribute process gradients along the device.
b) The Shallow Trench Isolation (STI): The DSM tech-
nologies use Shallow Trench Isolation (STI) for its accurate
dimension control when compared with LOCOS isolation [3].
STI is implemented in the form of trenches etched into the
wafer and filled with silicon dioxide to isolate the active area
of the transistors. Although STI provides some degree of latch-
up protection [4], this isolation technique induces mechanical
stress on the transistor and hence degrades its performance [5].
As shown in [6] and [7], this mechanical stress is highly
dependent on the layout style being used. To reduce the impact
of mechanical stress, the layout must be designed so that all
the transistors of the device are affected in the same way.
To keep a hard link between the electrical and the physical
view, the analog designer needs to have a precise estimation
of the impact of the layout on transistor characterization. This
makes the design more complex to be performed using the
manually traditional flow.
The automation of fundamental analog design steps, by
providing a two way-communication between the electrical
and the physical views, is extremely relevant for the success of
a project. Several works, such as [8], [9], [10] and [11], target
the automation of the layout generation of analog circuits.
To our knowledge, very few number of tools provide the
designer with a fast and accurate way to realize different
layouts for the same analogue atomic function. A remarkable
progress has been made by CIRANOVA [12] which develops
parametrized cells in Python Language, known as PyCells. Py-
Cells deliver physical views in OpenAccess [13] which is Ca-
dence standard and interoperable database. Many EDA startups
have invested in developing OpenAccess native applications.
OpenAccess is currently being pushed as a standard database
for the EDA industry. OpenAccess assures interoperability,
speed, usability. Yet, interoperability standards still need to
be agreed upon for this to be true.
In this paper, we will focus on developing a library of
analog basic building blocks called devices that supports
different layout styles (interdigitated, mirror, M2 module and
2D common centroid). We propose a methodology that allows
to size and bias a device and generate different layout styles
seamlessly. The idea is to well characterize the electrical and
the physical parameters of the device to meet functional and
robustness constraints [14]. Based on [15], sizing and biasing
operators are integrated in a loop with very fast nanometric
layout generation tool that allows to describe device layouts
in Python. The sizing operators propose sizes to the layout
generation tool. This in turn realizes the layout for a given
style. Then it computes physical sizes, stress effects as well
as layout dependent parameters. These are then fed back to the
sizing operator to be taken into account in the next iteration.
We show that the flow is very simple and achieves satisfactory
results in a matter of seconds. The advantage of the proposed
flow is to couple seamlessly and in a procedural manner both
transistor sizing and nanometric layout generation, with strong
The distances provided by the stack are :
• DMCI: distance from the middle diffusion contact till the
isolation edge.
• DMCG: distance from the middle diffusion contact till
the gate edge.
• DGG: distance between two successive gates. This is
equal to 2 × DMCG.
• DGI: distance from the edge of the end gate to the
isolation edge. This is equal to DMCI + DMCG.
Each distance has a method to query it in the stack object.
C. Layout Dependent Parameter Computation Methods
A dedicated Python API has been developed to describe
the device layout. In addition to the methods describing the
layout, three special methods have been developed to compute
the layout dependent parameters of the MOS transistor model.
The first one computes the area and perimeter of the source
and drain zones, the second one computes the stress effect
parameters as introduced in BSIM4 [16] and the third one
computes the capacitance of the routing wires.
D. Stress effect parameter computation
1) The stress effects for a transistor: In the BSIM4
model [16] the stress effect parameters are SA, SB, SD as
shown in Fig. 3.
• SA: Distance from the first left gate edge at the left end
of the stack till the isolation edge at the left end of the
stack. This is computed using:
SA = DGI + NBdummy × (Ldummy + DGG) (1)
where NBDummy is the number of dummies and
Ldummy is the dummy transistor length.
• SB: Distance from the first right gate edge at the right
end of the stack till the isolation edge at the right end of
the stack. This is computed using:
SB = DGI + NBDummy × (Ldummy + DGG) (2)
where NBDummy is the number of dummies and
Ldummy is the dummy transistor length.
• SD: Distance between two successive gates. This is set
equal to DGG.
Fig. 3. The stress effect parameters for the transistor.
W
L
W
L
Ids
Ids
����
M2M1
VIN+ VIN−
Fig. 4. Differential Pair Schematic View.
2) The stress effects for a Differential Pair: When consid-
ering the effects of mechanical stress on a differential pair, the
analysis differs significantly from the standalone transistor. For
the case of a standalone transistor, all the fingers belong to the
same device. On the other hand, a differential pair requires the
matching of two different transistors. The calculation of the
stress effects parameters becomes more complicated as it deals
with matched fingers from different transistors. In this case, the
calculation of stress parameters for a differential pair depends
on the layout style chosen for the differential pair. Fig. 5 shows
a differential pair consisting of transistors T1 and T2. In the
following, we discuss the stress effects calculations for the
layout styles: mirror and interdigitation.
Fig. 5. The stress effect parameters for the mirror style.
The stress effect parameters for differential pair can be
calculated as the transistor device by treating each transistor
(T1 and T2) separately using the equations (3)-(6).
InvSA,T1 =
(NFS−1)∑
i=0
δi
SA + 0.5 · Ldrawn + i · (SD + Ldrawn)
(3)
InvSB,T1 =
(NFS−1)∑
i=0
δi
SB + 0.5Ldrawn + i · (SD + Ldrawn)
(4)
SAeff =1
InvSA
(5)
SBeff =1
InvSB
(6)
We have improved the BSIM4 model to take into account
that the calculations of one transistor change according to
the position of its fingers. Note that the summation is carried
out on the range from 0 to NFS − 1. The presence of the
fingers of the other transistor alternating to the calculated
transistor’s fingers creates what we called holes . This requires
the introduction of a new parameter δ in the initial formula of
BSIM4. The parameter δ takes the value of 1 when pointing to
a finger of the transistor considered, and the value of 0 when
pointing to a finger of the other transistors.
The stress effects affect model parameters such as the
effective mobility µeff , the velocity saturation Vsat and the
Did intrinsic
target values ?
performances achieved+_
Sizing & Biasing
OPVS(Veg,Vb) W , Ve s
Choose Veg, Vd, Vg, Vb
Temp, Ids , L, gm0 target
START
END
∆I = I I
Style, W , L , Me ph
Parameters
yes
no
Layout Characterization
ph
e ph
ph
Ids, gm, gds, Cgs, Cgd, Cgb
W , L , Layout Dependent
OPIDS(W , L ,V , Layout Dependent Parameters) s
ds ds ds
Layout Generation
Fig. 6. Sizing and layout generation design flow
threshold voltage Vth [17], [18], [19]. To reflect the influence
of SAeff and SBeff , the parameter ”α” is defined as follows:
α =1
12SAeff
+ 12SBeff
(7)
Notice how stress effects, quantified by 1/α decrease by
increasing NF.
E. The design flow
Figure 6 illustrates how the operators and the layout method
are used to implement a device while respecting specifications.
In the first step, given the temperature, the biasing current
IDS , the overdrive gate voltage VEG, the drain voltage VD,
the gate voltage VG, the bulk voltage VB and the transistor
length L, the operator OPVS is used to compute the electrical
width We, the source voltage VS and the threshold voltage
VTH . The width We, length Le, number of fingers M and
layout style are given to the layout generator. Once the layout
is generated, the actual physical width Wph and length Lph
as well as the layout dependent parameters (diffusion zone,
stress and routing) are available. An accurate characterization,
including the actual physical realization, is performed using
the OPIDS operator that provides the actual IDS and the small
signal parameters for the purpose verification.
Here, we choose to take the gm as a specification for the
case of the differential pair. If the gm value does not meet the
specification, a loop is set to adjust the biasing current till the
specifications are achieved. After convergence, the final layout
is then realized.
Note that in the flow, all parameters varies, except the layout
style and the number of fingers that are kept invariant in this
loop.
III. RESULTS
A. Layout generation tool
In Fig. 7 we show the graphical user interface for the
layout generation tool CHAMS/Pharos. The upper left widget,
labeled Device Explorer shows the different layout parameters
set by the designer. The lower left message console prints the
values of all the layout dependent parameters. The script editor
in the lower right side of the GUI allows editing and executing
Python code.
Once the generation of the layout of the device is performed,
and thanks to the layout dependent parameter computation
methods, we have immediately the generic curves of the layout
dependent parameters versus NF as shown in Fig.8 and Fig.9:
a) Mirror style: Fig. 8 illustrates the generated curve 1/αversus NF for NMOS differential pair in mirror style in 65nm
technology with W = 6µm and L = 0.15µm.
Fig. 8. The stress effect 1/α parameter versus NF of the Mirror style.
b) Interdigitated style: Fig. 9 illustrates the generated
curve 1/α versus NF for NMOS differential pair in inter-
digitated style in 65nm technology with W = 6µm and L =
0.15µm. We can observe in Fig. 9 that, for the interdigitated
Fig. 9. The stress effect 1/α parameter versus NF of the interdigitated style.
technique, the stress effect affects the two transistors the same
way. While, for the mirror technique in Fig. 8, T1 transistor,
placed at the extremities of the stack, suffers a more significant
stress effect than transistor T2. This is due to the fact that
T1 is closer to the STI. So, the interdigitated technique in
the nanometer technology is more preferable unlike the older
technologies that prefer the mirror techniques.
B. Case study: Design of a Differential Pair
The goal is to design a differential pair in a CMOS 65nm
technology with a specified gm equals to 0.37 mS. The
assumption: VEG,VD, VG, VB and the transistor length L (set
Fig. 7. CHAMS/Pharos layout generation tool environment
Fig. 10. Differential Pair (interdigitated layout style) 65nm
Fig. 11. Differential Pair (M2 module layout style) 65nm
to 3.Lmin), are known. The design is performed for 4 different
layout styles with 4 fingers.
The results of the execution of the loop in Fig. 6 is
presented in the case of the four layout styles, Table II for
the interdigitated style, Table III for the mirror, Table IV
for the M2-module style and Table V for the 2D common-
centroid. The Iter parameter is the number of iteration in the
loop, Wph is the transistor width, IDS is the drain current
biasing the transistor, gm is the transistor transconductance
and SAeff , SBeff and 1/α are the stress parameters defined
by the BSIM4 model. The corresponding layouts are presented
in Fig. 10 (interdigitated), Fig. 11 (M2 module), Fig. 12 (2D
common-centroid) and Fig. 13 (mirror) respectively.
Fig. 12. Differential Pair (2D common centroid layout style) 65nm
Fig. 13. Differential Pair (mirror layout style) 65nm
Let us examine Table V for the differential pair with 2D
Common Centroid style . After all the loop iterations to get
the required gm, we can get many information about the
electrical and the physical parameters such as the electrical
intrinsic capacitances: Cgs = 2.39fF , Cds = 1.16fF ,
Cgd = 0.558fF , Cgb = 0.37fF , etc ... Also the physical
parasitic routing capacitances: Cgph= 0.0581fF ,Cdph
=0.0937fF ,Csph
= 0.03.89fF , etc ... related to the net gate
TABLE IIRESULTS 65NM INTERDIGITATED (T2 TRANSISTOR)
Iter Wph IDS gm SAeff SBeff 1/α(µm) (µA) (mS) (µm) (µm) (µm−1)
1 0.9 29.37 0.245 2.22 0.81 0.842
2 0.93 29.98 0.25 2.22 0.81 0.842
27 1.43 43.64 0.364 2.22 0.81 0.842
28 1.45 44.19 0.369 2.22 0.81 0.842
TABLE IIIRESULTS 65NM MIRROR (T2 TRANSISTOR)
Iter Wph IDS gm SAeff SBeff 1/α(µm) (µA) (mS) (µm) (µm) (µm−1)
1 0.9 29.60 0.247 1.99 1.99 0.5025
2 0.93 30.30 0.253 1.99 1.99 0.5025
26 1.41 43.62 0.363 1.99 1.99 0.5025
27 1.43 44.18 0.368 1.99 1.99 0.5025
TABLE IVRESULTS 65NM M2 (T2 TRANSISTOR)
Iter Wph IDS gm SAeff SBeff 1/α(µm) (µA) (mS) (µm) (µm) (µm−1)
1 0.9 28.24 0.236 0.48 0.48 2.08
2 0.94 29.07 0.243 0.48 0.48 2.08
29 1.485 43.18 0.362 0.48 0.48 2.08
30 1.505 43.74 0.366 0.48 0.48 2.08
TABLE VRESULTS 65NM 2D-COMMONCENTROID (T2 TRANSISTOR)
Iter Wph IDS gm SAeff SBeff 1/α(µm) (µA) (mS) (µm) (µm) (µm−1)
1 0.9 28.92 0.241 1.59 0.48 1.356
2 0.93 29.52 0.247 1.59 0.48 1.356
28 1.45 43.43 0.363 1.59 0.48 1.356
29 1.47 43.96 0.368 1.59 0.48 1.356
’G’, drain ’D’ and source ’S’ respectively. We can calculate
similarly the transition frequency defined as:
Ft =gm
2Π(Cgs + Cgd + Cgb + Cgph+ Csph
)= 17.6GHz
The overall computation time for the optimization loop is
around 6s (1s for the sizing, 5s for layout generation). The
average number of iterations is less than 30.
We conclude from the tables that the target gm can be
achieved under the influence of stress effect for the different
layout styles. This has to be compromised with matching
requirements in order to choose the final layout style.
IV. CONCLUSION
In this paper, a new method for developing smart
parametrized generators for analogue devices has been pre-
sented. The interaction between the transistor sizing and the
layout generation of the device has been illustrated in the
case of the differential pair. Four different layout styles have
been compared. The tight coupling between the transistor
sizing and the layout generation has been used to accurately
characterize a design. The proposed method allows the de-
signer to select the most convenient device layout style for
given specifications. The results showed the efficiency of the
proposed method. As a future work, the proposed method
will be extended to allow a seamless coupling between circuit
level sizing and layout generation, taking into account intrinsic
device performance.
REFERENCES
[1] B. Razavi, Design of Analog CMOS Integrated Circuits. McGraw-Hill,2000.
[2] Ravindranath Naiknaware et al, “Automated Hierarchical CMOS AnalogCircuit Stack Generation with Intramodule Connectivity and MatchingConsiderations.” J. Solid-State Circuits, vol. 34, no. 3, pp. 304–317,March 1999.
[3] D. Andriukaitis et al, “LOCOS CMOS process simulation,” in Conf.
Information Technology Interfaces, Catvat, Croatia, June 2006, pp. 489–494.
[4] W.L. Goh et al, “A comprehensive geometrical and biasing analysis forlatchup in 0.18-µm CoSI2 STI CMOS structure.” jssc, vol. 48, pp.2109–2114, 2004.
[5] ——, “Latchup characterization of 0.18-micron sti cobalt silicided teststructures.” Microelectronics Journal, vol. 32, no. 9, pp. 725–731,September 2001.
[6] M. Ranjan et al, “Fast, layout-inclusive analog circuit synthesis usingpre-compiled parasitic-aware symbolic performance models,” in Proc.
Design, Automation and Test in Europe Conf., Paris, France, February2004, pp. 604–609.
[7] V. Joshi et al, “Stress aware layout optimization,” in Proc. Int. Symp.
on Physical design, Portland, Oregon, April 2008, pp. 168–174.[8] D. Stefanovic, M. Kayal, M. Pastre, and V. B. Litovski, “Procedural
analog design (pad) tool,” in Proceedings of the 4th International
Symposium on Quality Electronic Design, ser. ISQED ’03. Washington,DC, USA: IEEE Computer Society, 2003, pp. 313–. [Online]. Available:http://portal.acm.org/citation.cfm?id=851002.855973
[9] L. Lewyn and N. Williams, “Is a new paradigm for nanoscale analogcmos design needed?” Proceedings of the IEEE, vol. 99, no. 1, pp. 3–6, Jan. 2011.
[10] N. Jangkrajarng, S. Bhattacharya, R. Hartono, and C. J. R. Shi, “Iprail–intellectual property reuse-based analog ic layout automation,” Integra-
tion, the VLSI Journal, vol. 36, no. 4, pp. 237 – 262, 2003, analog andMixed-signal IC Design and Design Methodologies.
[11] R. Castro-Lopez, O. Guerra, E. Roca, and F. Fernandez, “An integratedlayout-synthesis approach for analog ics,” IEEE Trans. Computer-Aided
Design, vol. 27, no. 7, pp. 1179–1189, Jul. 2008.[12] “http://www.ciranova.com/.”[13] “http://www.cadence.com/.”[14] T. Massier, H. Graeb, and U. Schlichtmann, “The Sizing Rules
Method for CMOS and Bipolar Analog Integrated Circuit Synthesis,”in Computer-Aided Design of Integrated Circuits and Systems, IEEE
Transactions on, vol. 27, no. 12, Dec. 2008, pp. 2209 –2222.[15] F. Javid, R. Iskander, and M.-M. Louerat, “Simulation-Based Hierar-
chical Sizing and Biasing of Analog Firm IPs,” IEEE International
Behavioral Modeling and Simulation Conference, pp. 43–48, Sep. 2009.[16] W. Liu, MOSFET Models for SPICE Simulation including BSIM3v3 and
BSIM4. Wiley Interscience, 2001.[17] L. Yang et al, “Simulation of layout-dependent STI stress and its impact
on circuit performance,” in Int. Conf. on Simulation of Semiconductor
Processes and Devices, San Diego, USA, September 2009, pp. 281–284.[18] H. Tsuno et al, “Advanced analysis and modeling of MOSFET char-
acteristic fluctuation caused by layout variation,” in VLSI Technology
Symposium, Kyoto, Japan, June 2007, pp. 204 –205.[19] V. Moroz et al, “The impact of layout on stress-enhanced transistor
performance,” in Int. Conf. on Simulation of Semiconductor Processes
and Devices, Tokyo, Japan, September 2005, pp. 143 – 146.